Geoscience Reference
In-Depth Information
42 CHAPTER 4. CO-TRAINING
BIBLIOGRAPHICAL NOTES
Co-Training was proposed by Blum and Mitchell [
22
,
129
]. For simplicity, the algorithm presented
here is slightly different from the original version. Further theoretical analysis of Co-Training can be
found in [
12
,
10
,
53
]. Co-training has been applied to many tasks. For examples, see [
41
] and [
93
]on
named entity classification in text processing. There are also many variants of co-training, including
the Co-EM algorithm [
134
], single view [
77
,
38
], single-view multiple-learner Democratic Co-
learning algorithm [
201
], Tri-Training [
206
], Canonical Correlation Analysis [
204
] and relaxation
of the conditional independence assumption [
92
].
Multiview learning was proposed as early as in [
56
]. It has been applied to semi-supervised
regression [
25
,
159
], and the more challenging problem of classification with structured outputs [
24
,
26
]. Some theoretical analysis on the value of agreement among multiple learners can be found
in [
65
,
110
,
154
,
193
].